Codex 4.8 vs Claude, Hermes OS, Ecom Stack — AI Daily Jun 1

528 messages · 78 active members

528
messages
78
active members
@anonymous, @jasonakatiff, @rmktg
top contributors

Overview

Today's discussion split between agent infrastructure and ecom execution. @jcartu led the harness wars, arguing the harness matters as much as the model — recommending OpenCode + OMO + hindsight over vanilla Claude Code, which he blames for Opus 4.8's 'compliance-whore' behavior. His team is building KAGED (kaged.dev). @namtalks evangelized the 'Hermes OS' approach: treating the agent's governance/rules folder as the brain of the business, with structured workflows that scale across any business model. MiniMax at 20-30tps was dismissed as unusable for agentic work. On the creative side, anonymous solved Veo 3.1's audio quirks by switching from text to JSON prompts, anonymous compared SD2 vs Grok Imagine 1.5 (SD2 wins cinema, Grok wins speed), and Sora beat Veo3/Seedance on lip sync. anonymous refined creative briefs to one concept with multiple angles, while @jasonakatiff shared his mass ad launcher inspired by Akatiff's tool. Bulk launching = many adsets on $500 tests, not big budgets. Ecom got a deep thread when a newcomer asked about dropshipping: veterans warned against Zendrop's wrong-product shipments at scale, recommending early migration to private 1688/Ali suppliers. Adjacent threads covered jailbreaking Claude Code for aggressive ad copy, Fin.ai vs Claude+Helpscout for support automation, ugly 90s landers outconverting polished ones, weekly creative KPIs with PIPs, and Wispr Flow outages disrupting voice workflows.

Topics

Agent Harness Wars: Codex 4.8, OpenCode + OMO, and KAGED18 msgs

@jcartu argued the harness drives behavior — blaming Claude Code for Opus 4.8's over-cautious compliance and recommending OpenCode with OMO + hindsight. His team is building KAGED (kaged.dev) with his CTO leaning toward Pi. Codex 4.8 was also flagged as overtaking Claude for code implementation work.

Hermes as a Business OS: Governance Folders and Workflow Design14 msgs

@namtalks pitched Hermes paired with a properly structured OS as the agent backbone for any business — treating the 'governance' folder as a must-read brain. @robinroy already runs Hermes against his Telegram via OpenClaw. Nous Research's NVIDIA RTX Spark partnership for Hermes also dropped today.

Veo 3.1, Sora, and AI Ad Creative Workflows33 msgs

anonymous fixed Veo 3.1 ignoring 'no audio' instructions by switching to JSON prompts. Sora beat Veo3/Seedance on lip sync; SD2 vs Grok Imagine 1.5 split on cinema vs speed. anonymous restructured briefs around one concept with multiple messaging angles, and @jasonakatiff shared his drag-and-drop mass ad launcher with page/pixel/location targeting.

Dropshipping Stack: Zendrop Warnings & Private Suppliers25 msgs

Newcomer @realcrischico kicked off a deep thread on ecom entry — Evolve recommended as the course, Dropified vs Zendrop debated. Veterans warned Zendrop ships wrong products at scale with sketchy gig-economy last-mile delivery; use them to learn, then graduate to private 1688/Ali suppliers ASAP for the dropship→brand pipeline.

AI Copy Guardrails, Support Automation, and Bulk Ad Launching30 msgs

Claude 4.8 increasingly refuses aggressive ad copy — workaround: use Claude Code (not Desktop) and explicitly strip compliance guardrails. For support: Fin.ai at $1/outcome vs leaner Claude + Helpscout with KB + read-only DB access and one human reviewer. Bulk Meta API launching via Hermes = $500 tests across 200 adsets, not big budgets.

Key Takeaways

  • The harness matters as much as the model — try OpenCode + OMO or KAGED before blaming Opus 4.8 for over-compliance.
  • When Veo 3.1 ignores audio instructions, switch from text prompts to JSON-formatted prompts; Sora still wins on lip sync.
  • Skip Zendrop for scale — wrong-product shipments and gig-economy last-mile delivery break trust; go private 1688/Ali suppliers early.
  • Use Claude Code (not Desktop) and explicitly instruct it to strip compliance guardrails for aggressive ad copy generation.
  • Bulk ad launching = many adsets, not big budgets — run $500 tests across 200 variations in a CBO and let Meta pick winners.

Hot Threads

@realcrischicostarted

Getting into ecom/dropshipping — best course, Dropified vs Zendrop, supplier strategy

25 replies8 participants
@jcartustarted

Harness choice drives Opus 4.8's compliance behavior — OpenCode + OMO recommended, KAGED in development

10 replies3 participants
@namtalksstarted

Hermes + a properly structured OS as the agent backbone for any business model

9 replies5 participants

Linked Items

88.5% Performance Boost: The Secret Behind Claude Code's Success

x.com
@GuruTime

Unable to analyze - Instagram content not accessible

instagram.com
@iannagy

Student's $2 LED Shows When AI Agents Are Working - Built in 15 Minutes

x.com
@anonymous

AI Ad Lab: Create 1000+ Ads/Month Like $150M+ Agencies Do

skool.com
@roki1122

1.65 GB folder on MEGA

mega.nz
@anonymous